Recreate the work.
Give models tasks that reflect the tools, context, and decisions of healthcare.
Frontier intelligence. Grounded in care.
Aligning AI
with patient care.
We build healthcare environments where frontier models can learn from expert judgment—and be evaluated against the demands of real care.
Partner with usFor frontier labs & healthcare AI buildersOur approach
01 Our conviction
The frontier of AI should
move patient care forward.
Medical knowledge is only one part of healthcare. The work also demands judgment: navigating incomplete records, coordinating across systems, recognizing uncertainty, and knowing when a clinician needs to take over.
Our mission is to help models earn their place in that work. We turn healthcare expertise into environments for learning and evaluation, with patient care as the reason to improve.
02 Our approach
Make expert judgment
part of how models learn.
We build tasks around healthcare workflows, with success criteria defined by people who do the work. Evaluation examines the path a model takes alongside the answer it gives.
Examine the behavior.
Use expert-defined criteria to identify mistakes and gaps in performance.
Inform better training.
Turn expert feedback into signals for post-training and subsequent evaluation.
03 The work today
Grounded in the details
that shape care.
Our current work focuses on healthcare administration and radiology: different workflows, connected by the need for reliable judgment.
Healthcare administration
Tasks spanning health records, payer portals, and office applications, written by US healthcare administration experts. Our goal is to help models navigate the administrative work that stands between patients and care.
Radiology reporting
Reporting tasks written and graded by physicians. We examine model performance against expert-defined criteria for correctness, connecting clinical knowledge to the demands of reporting.
04 The future we want
Healthcare should own
the intelligence it relies on.
We envision models shaped by a health system’s expertise, evaluated for its workflows, and able to operate within its own infrastructure. Patient information should stay under the institution’s control.
That future includes settings where cloud access is restricted—or unavailable. Our work in environments and expert evaluation is a foundation for that ambition: helping healthcare build the evidence and learning signals its models need.
Local expertiseInstitutional controlAccountability to care
Our direction: healthcare-owned models and private deployment. Our offering today: expert-built environments and evaluation.
05 Behind the work
Built by people with roots in
science, medicine, and AI.
EducationIndian Institute
of ScienceBiology
of ScienceBiology
Research experienceHarvard
Medical SchoolMedical research
Medical SchoolMedical research
Company buildingEntrepreneurs
FirstBuilding out of EF
FirstBuilding out of EF
06 Work with us
Build toward that future.
For frontier labs and healthcare AI builders
Partner with us on healthcare training environments and expert-led evaluation. Bring a target workflow, a model capability you want to improve, or a failure mode you need to understand.
For clinicians and healthcare organisations
We want to learn where current models fall short in your work, what good performance looks like, and what evidence you need before considering their use. Contact us to explore a collaboration.
What can we work on today?
Expert-defined healthcare tasks, grading criteria, and training and evaluation material, including containerized tasks in Harbor format. We agree on the workflow, scope, and deliverables for each engagement.
Do you deploy models inside hospitals today?
Private and on-premises models are part of our longer-term direction. Our current offering focuses on healthcare environments and expert evaluation; this website does not announce an available hospital deployment product.